Available online at ScienceDirect. Procedia Computer Science 60 (2015 )

Transcription

1 Available online at ScienceDirect Procedia Computer Science 60 (2015 ) th International Conference on Knowledge Based and Intelligent Information and Engineering Systems Tag-based Chat Support System to Remind Users of Contents of Past Conversations Rina Tanaka a, Junko Itou b,, Jun Munemori b a Graduate School of Systems Engineering, Wakayama University, 930 Sakaedani Wakayama City, , Japan b Faculty of Systems Engineering, Wakayama University, 930 Sakaedani Wakayama City, , Japan Abstract In this article, we propose a chat system that helps users remember and resume past conversations by using tags. In computermediated communication such as online chat, it is often difficult in communication to continue conversations regarding issues that have been discussed in the past because they may have forgotten the contents of the issue. Our system adds tags for each chat logs based on words that were used in the chat and displays the tags when users restart the interrupted chat. As a result of three experiments with the proposed system, it revealed that display of tags is useful to remind users of contents of past conversations. c 2015 The Authors. Published by by Elsevier B.V. B.V. This is an open access article under the CC BY-NC-ND license Peer-review ( under responsibility of KES International. Peer-review under responsibility of KES International Keywords: Chat, Tag, Topic, Reminder, morphological analysis. 1. Introduction Various tools have been used as a means to communicate with persons who are in remote places, such as letters, telephone, bulletin boards, chat, and instant messenger. In recent years, many people use free communication applications such as Skype 1 and LINE 2 with the development of internet communication network and smartphones. In these applications, users are able to exchange messages including text and images for free. In computer-mediated synchronous communication such as online chat, conversational contents are often mixed because users typically discuss several topics in the same session. Users are also apt to forget what they were discussing with their online interlocutors as they are frequently compelled to interrupt their conversations. Therefore, it is often difficult in such communication to continue conversations regarding issues that have been discussed in the past. In order to recall the topics that they may have forgotten, users need information regarding these. However, at present, users need to manually save chat logs and re-read them, all the while being able to recall the nuanced transitions between topics, in order to recommence online conversations. That is to say, they cannot easily obtain clues that can help them remember past online conversation topics. Corresponding author. Tel.: ; fax: address: itou@sys.wakayama-u.ac.jp The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of KES International doi: /j.procs

2 892 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) In this paper, we propose a chat system that helps users remember and resume past conversations by using tags. Our target is an informal chat instead of a conference such as taking the minutes. Our system shows to the user words associated with an ongoing chat topic on the screen at any given time. These words are nouns extracted from chat logs by morphological analysis. The user can select multiple words as tags that represent the topic, and help remind him/her of the content of the relevant conversation later. A few related systems focusing on online chat have been proposed. One such proposed system can glean the topic of an ongoing conversation by analyzing chat logs 3, but cannot help users remember past conversational topics. The other system does not record online chat logs in chronological order, but divides them into topics within the ambit of a given theme and those beyond it 5. This system controls the speed at which messages are deleted from logs according to their levels of importance. In this system, the user can choose to associate any block of text with any column representing a theme considered by him/her to be relevant. Therefore, this system does not assume that users re-read past logs to recall the contents of past chats. By contrast, our system records the switch timing of conversational topics in order for users to grasp transitions between topics. This paper is organized as follows: in section 2, we will describe the problems about existing systems to support chat communication focusing on chat topics. In section 3, we explain how our system supports to remind past topics using tags and how the associated words are extract from a chat log. We will show the experimental result on effectiveness recording and displaying tags in section 4. Finally, we will discuss some conclusions and our future steps in section Related work on chat communication focusing on conversational topics Opportunities of computer mediated synchronous communication are increasing and many remote communication systems have already been proposed. One of existing typical chat services are Skype 1 and LINE 2. Skype allows users to register their icons, user names, and notifications indicating their statuses or moods. A chat history is shown once the user logs out. Users are also allowed to restart conversations with chat partners. LINE is a communication application that allows text and multimedia image communication. A message is shown in a balloonshaped textbox, along with a timestamp and message read mark. New messages are inserted under old ones in the same window. In these services, it is possible to save each conversation as a text file. However, only the messages and their transmission times are recorded. Thus, users need all conversation logs while relying on their memory to determine where the relevant conversation was interrupted, or what they were talking about at the time. Kawabata et al. proposed a system that extracts chat topics from a chat room using a history of messages 3. This system presents suitable words for an ongoing conversation in a chat room in order to introduce other users, who have not yet joined the conversation, to the chat contents. In this system, a chat log saved every five minutes is divided into three parts - a current conversation (i.e., the conversation ongoing at the time), a last-minute conversation, and a past conversation. This system extracts nouns from each log using the Japanese morphological analyzer MeCab 4. The log for a current conversation is used to obtain the characteristics of the conversation in the relevant chat room. Since last-minute topics tend to shift to a current topic, the last-minute log is analyzed to extract the characteristics of the current topic. For users who have conversed about specific subjects in the past, the system considers it likely that they will be chatting about similar issues at any given time. In order to incorporate the features of user participation in the conversation at this stage, the system extracts nouns from chat logs and gleans the conversational theme at the time. Users intentions are not reflected in this classification since the results of analysis in this system are only used to classify conversations into broad topics such as food, hobbies, politics, and economic. Moreover, Kawabata et al. s system does not help remind users of past chat topics. In all prevalent chat and instant messaging services, users can simultaneously pursue multiple topics in a conversation. At the same time, it is sometimes difficult to accurately grasp the flow of the conversation when multiple topics are being discussed, especially if several users are participating in the conversation. Collective Kairos Chat 5 is a chat support system where users can determine the degree of importance of each message. This system allows users to delete chat messages from the log at different speeds. The chat screen in the system has three columns, and messages are divided in accordance with their degree of importance as determined by all users. When chat participants have offtopic conversations, they assign the relevant messages to the column containing relatively less important messages. The log of the column containing highly important messages flows slowly, whereas the logs of the less important

3 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) columns flow more quickly. As a result, important messages germane to a given theme are displayed for longer, and messages and conversations tangential to the theme are shown briefly in order for users to obtain chat logs as the collective memory of the discussion reflecting the preferences of all participants. The criterion for the importance of a message is whether it is associated with a given theme. Hence, Collective Kairos Chat does not cater to situations where multiple conversational topics are spanned in a short time. In this paper, we propose a support system to facilitate checking the contents of the past issue and to restart chatting on the issue by tagging and registering topics switch timing. 3. Proposed chat support system 3.1. Goal Our goal is to implement a chat system that supports the recommencement of past chat topics by tagging chat logs. Users can remind themselves of the contents of past chats by checking the tags associated with them as well as records of topics switch timing. It is assumed that the proposed system is used in frank conversation by college students about every other week, such as a simple meeting or talk several times a month, rather than a formal meeting, such as taking the minutes. In current chat systems, it is difficult to quickly grasp the content of past conversations because users need to read entire chat logs. In order to solve this problem, our proposed system uses tags based on chat conversations. The design policy of our system is as follows: 1. A mechanism to help users easily grasp the contents of past chat conversations In order for users to continue discussion on a past chat topic, information regarding the topic, such as the relevant conversation log, is needed. It takes time and effort to keep track of all logs; hence, our system registers nouns from chat conversation as tags. Users can fleetingly confirm the contents of a conversation by checking the tags associated by the system with it during the chat. 2. Recording the transition between topics by user eyes The proposed system only presents candidate tags to the user, who ultimately decides which tags to assign to a given conversation. Moreover, users can register the timing at any point when they feel the topic of the conversation has changed System structure System overview Our chat system consists of two clients connected to a network. These clients have candidate tag extraction functions. Fig.1 shows the system s structure. Our system facilitates socket communication between two computers. A client computer sends a request to a receiver computer. The receiver computer receives the request, and chats with the client until disconnection. Chat logs are saved to a text file for candidate tag extraction. The log for each message contains the message, its time of transmission, selected tags, and topics switching time. When a user starts the system, a window comes up as shown in Fig.2. An input form is located at the top of the screen. The user enters text here, and clicks the submit button or presses the Enter key to send a message to the other user. s are shown below the input form, and new messages are added under already displayed messages. Tags that users have registered for past chats are shown to the right of the chat screen. Fig.2 shows five tags. The number of tags is not limited. In order to register a tag for an ongoing chat, the user clicks the tag registration button at the bottom of the screen. The system then runs a tag extraction function and shows candidate tags to the user. The user clicks a candidate tag, which causes the color of the candidate tag button to change, indicating that the candidate has been recorded as a tag for the conversation. There is no upper limit on the number of tags that can be displayed. If a new candidate tag is required as the chat proceeds, users can register them by clicking the tag registration button. The function extracts new candidate tags, shows them to the user, and adds the chosen candidates to the list. Finally, when the user presses a topic switch button at the bottom of the screen, the time at which the button is pressed is recorded.

4 894 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) Extracting candidate tag Extracting candidate tag Exchange Candidate tag Client Client Candidate tag Fig. 1. System structure Extracting candidate tags The tag extraction function operates on each client computer. The function first analyzes a chat log stored on each client s computer using MeCab 4. MeCab is the Japanese language morphological analyzer. The log data are divided into the parts of speech, and only nouns are used as candidate tags. The extracted nouns include overlapping words or words that do not make sense, e.g., a number such as a date or a price, a pronoun, a suffix, or an emoticon. The function eliminates nouns that are not used as tags. Thus, certain nouns are selected as candidate tags. Candidate tags are allocated in temporal sequence, i.e. the word is appeared in the current chat text Registration of tags When the user clicks the tag registration button, the system runs the tag extraction function using the chat log accumulated for the given conversation until then, and extracts candidate tags by the method described in Section The client user can see candidate tags at the bottom of the screen. If the user clicks on a candidate tag, it is chosen as a tag for the conversation, and this is signified by a change in the color of the tag. The screen after the tags are registered is shown in Fig.2. In the example in the figure, two candidate tags were selected. The time clicked and the tag are recorded in the conversation log, and the registered will be shown in subsequent chats. The target of our system is frank conversation for minutes a few times a month, so a formal meeting required to record the exact proceedings is not assumed. Additionally, our system aims to provide an opportunity to recall the contents of the topic and it does not support that remind the details of talk. Therefore the registered tag are aligned in the order they appeared in the chat text for each topic not in the order they registered. 4. Comparison experiment 4.1. Experiment outline We implemented the system described in Section 3 for a comparison experiment to determine whether the tags proposed in our system help remind users of past conversational topics, and whether the operations to display and register these tags interfere with the chat.

5 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) Text input form Contents of past conversation Tag registration Topic switch Submit registered tags for past chat snow graduation thesis movie travel Kyoto last time graduation thesis movie talking graduation travel Candidate tags in peace Fig. 2. Chat window. The subjects in our experiment were 12 college students, nine males and three females, divided into six pairs. All subjects were accustomed to handling keyboard input and were familiar with chat conversations. The subjects were asked to chat while sitting in the same room, with any two subjects separated by more than three meters in order to prevent directly conversations. The participants used two types of systems, the comparison system and the proposed system. The comparison system removed the only display of tags on the last topic from the proposed system. Other related functions such as chat, the tag registration function, and the interface in the system were similar to those of the proposed system. The chat screen of the comparison system is shown in Fig.3. In both systems, the participants were imposed no restrictions in using these systems. We conducted three experiments at intervals of one day or more. In the first experiment, the participants used the comparison system, and we collected data for the second and the third experiments. In the second experiment, three pairs of participants used the proposed system while the other three pairs used the comparison system considering the influence of order effects. In the third experiment, the participants used the other system in the second experiment. At the conclusion of the each experiment, we asked the subjects to answer a few questions. We also carried out a survey regarding the tags registered by the participants in the first experiment more than a week after it. We describe the contents of our instruction in the each experiment. The first experiment The first chat experiment was performed to collect data regarding the last topic to use in subsequent experiments. 1. We explained the operation of the system to the participants, and asked them to press the topic switch button if they felt that the topic of the conversation had changed. Since there was no log for a previous with respect to the first chat, it was not possible to display tags from the previous conversation for the first chat. For this reason, the participants used the comparison system. 2. We proposed themes for conversations such as a hobby, a TV show, travel, their hometowns, and progress in their academic theses in case participants found it difficult to generate one. 3. We instructed each pair to chat for 15 minutes using the system, and register tags during the chat. The tags were selected according to whether they would be useful to remember the topic of conversation later.

6 896 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) Text input form Translation Hello. How are you? Fine. Recently, I feel the air is cold. I think so. I m hard to get out of bed each morning from cold : ] Submit Tag registration Topic switch fine recently air morning snow graduation thesis schedule one another movie this time Candidate tags Fig. 3. The comparison system. Second and third experiments 1. We explained the operation of the system to the participants. In particular, we emphasized the feature of the proposed system whereby it could show tags chosen in previous conversations. 2. We asked the participants to recall the topic of the conversation from the first experiment and chat for 15 minutes using our system. 3. We instructed each pair to register the tags they thought they would need to remember topics later during the chat Experimental result questionnaire result We denote experiments that used the proposed system by exp p and those that used the comparison system by exp c. The results of subjects responses to our questionnaire are listed in Table 1. The represent mean values on a five-point scale. The averages were rounded to the second decimal place. In the questionnaire, which was posed more than a week after the first experiment, the participants answered about items (ix) and (x). At first, we discuss restart a conversation about the last topic. Questionnaire items (i) and (ii) related to the topic of the previous conversation and its recollection. As shown in Table 1, participants assignment of scores for item (i) for our system exceeded those for the comparison system. However, the variance in the results of item (ii) for the proposed system was higher than that for the comparison system. Hence, a significant difference in neither item could be obtained between the comparison system and our proposed system. The participant who assigned an especially low score for item (ii) reported feeling that that he/she had exhausted the topic, and could not find any more things to talk about. However, his/her chat partner reported being able to resume the topics of past conversations. Hence, it seems that the conversation recollection differed greatly between members of the same pair.

7 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) Table 1. The average values of the questionnaire result. Questionnaire item exp p exp c (i) The system remind me of the topic of my last conversation (ii) I was able to recommence conversing about the last topic (iii) I was able to stump about how to say (iv) The contents of tags were appropriate (v) The registration of tags was not troublesome (vi) Seeing tags helped me restart a conversation about the last topic (vii) Seeing tags helped me remember the last topic (viii) The way the tags were displayed was easy to see (ix) The way the tags were displayed reminded me of the content of the last chat (x) The way the tags were displayed reminded me of timing when topics were switched Evaluation value: 1:strongly disagree, 2:disagree, 3:neither, 4:agree, 5:strongly agree We asked subjects about the candidate tags displayed on the chat screen in our system in questionnaire items (iv) and (v). A clear difference between the two systems was not observed. Since the tag registration operation was identical in both systems, we conclude that other differences between the systems did not affect the responses to item (v). As mentioned, our experiments were conducted at least one day apart. Hence, it is possible that some participants memorized the contents of a conversation such that they could remember these without being shown tags. In fact, in the questionnaire, we obtained a score of 4.9 for item (ix). This value was higher than that for item (i). Hence, in future work, we intend to conduct experiments at least one week apart. Questionnaire item (vi) attempted to determine whether the displaying of tags had helped participants recommence a previous conversation. The score for this item was 3.6 with a variance of 2.1. Therefore, there was a significant divide in opinion among the subjects regarding whether the tags were helpful in restarting past conversations. Given the results of items (i), (vi), and (ix), we think it necessary to provide further support for talk in detail about the topic. The variance value for item (vii) was the largest of all items at 2.3. Three participants selected the value 1 and two selected 5. The participants who assigned a low score to item (vii) tended to assign a low score to item (vi) and a high score to item (iii). We think this was the case because these participants did not need tags to remember and recommence past conversations. Some participants reported finding the registration of tags inconvenient because of restrictions on tagging. To solve these problems, we need to implement an input form that allows users to manually enter any word as a tag, and not merely nouns extracted from the conversation log The number of pressed the tag registration button and the topic switch button Table 2 shows the number of times the tag registration button and the topic switch button were pressed for each participant. The participant p1 - p6 used the proposed system in the second experiment and the participant p7 - p12 used the comparison system in the second experiment. The participant p1 and p2, p3 and p4, p5 and p6 were pair of chat. For the first experiments that tags on the previous topic were not displayed and the second or the third experiment that registered tags were displayed, a significant difference in the number of registered tags was observed by the Wilcoxon signed rank test. For the number of pressed topic switch button, the results of the Wilcoxon signed rank test indicated no significant difference between the two cases. The tag number per one topic are calculated based on the number of switching topics and registered tags. The topic numbers that the participants talked in the experiments are the number of pressed topic switch button plus 1. By dividing the registered tag number in this number, the tag number of per topic is calculated. Additionally, we asked the participants to answer two questions; 1. How many the ideal tag numbers displayed at the right side for one topic? 2. Was the displayed tag numbers adequate? The answers are shown in the right columns of Table 2 Referring to Table 2, seven participants tag number per one topic are less than the number of ideal tag numbers. Nine participants felt the number of tags displayed on the right side is proper, and none of participants felt the displayed tag number is poor. In contrast, three participants felt the tag number is too much. The number of tags that these

8 898 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) Table 2. The number of registered tags and switching topics. User Registered tags Switching topics Tag number Ideal tag number Displayed ID First Second Third First Second Third per one topic per one topic tag number p too many p adequate p adequate p adequate p adequate p too many p adequate p adequate p adequate p adequate p adequate p too many Avg participants actually registered were seven to eight, and the registered tag numbers per one topic of two participants were below the ideal tag number. One of the reason for this result is due to the square size and the display position of the tag. Furthermore, the number of tags attached to one of the experiment were from 3 to 35. The registered tag numbers differ according to each participant even participants of a same pair. Our system allows each users to register tags as needed so it is considered that the variability of the tag numbers is the result reflected the user s intentions successfully. We discuss the number of pressed topic switch button. Table 2 shows a list of the number of times the topic switch button was pressed in each chat. There is a large variation in the same pair of participants for the timing of topic switching. Since the timing and the number can be changed by the subjectivity of the participants, the functions that each user can freely register the timing is useful. On the other hand, the participants answered in the free description that they could not feel free to press the topic switch button because the button and the input form is located away, and they had forgotten to press because they were distracted by chat. Therefore, it is necessary to examine the the arrangement of topic switch button. 5. Conclusion In this paper, we propose a chat support system to help users restart a conversational topic from past chats. This system checks the contents of past conversations by tagging them based on their contents. The system applies Japanese morphological analysis to chat logs and displays nouns used in the chat as candidate tags. Users can then select and register tags that they deem helpful in remembering chat contents. Tags registered in previous chats are displayed on the right side of the chat screen. We performed an experiment to compare our proposed system, which displays tags associated with past topics, with a chat system that did not do so. The results of a questionnaire filled out by subjects of our experiments showed that the proposed system helped remind users of topics of the past conversations by showing tags from past chat logs. The evaluation score for the question I was able to recommence conversing about the last topic was relatively high, but the variance was large. This result suggests that further support is required for restart of the past topic due to individual differences. In future work, we plan the following: a review of the experimental period, support for conversations regarding past topics, implementation of a manual input form, and the improvement of the interface. Acknowledgements This work was supported by JSPS KAKENHI Grant Number

9 Rina Tanaka et al. / Procedia Computer Science 60 ( 2015 ) References 1. Skype: (Access on March 25, 2015) 2. LINE: (Access on March 25, 2015) 3. Kawabata, T., Satou T., Murayama T. and Tada Y.: The construction of the topic extraction system Base on statement history in the chat room Proceedings of the 77th national convention of Information Processing Society of Japan, vol. 72, no.1, pp , 2010 (In Japanese). 4. MeCab: Yet Another Part-of-Speech and Morphological Analyzer: (Access on March 25, 2015) 5. Ogura, K., Matsumoto Y., Yamauchi Y. and Nishimoto K.: Kairos Chat: a novel text-based chat system that has multiple streams of time Proceedings of the CHI 10 Extended Abstracts on Human Factors in Computing Systems, pp , 2010.